SwallowCare

The app asks how it went, not how much.

Company

SwallowCare × Georgia Tech MS-HCI

My Role

Lead Product Designer

Tools

Figma · Qualtrics · UserTesting

Timeline

Aug – Dec 2025

Description

A daily-adherence app to bridge the gap between clinic visits for people with dysphagia, a swallowing disorder where skipping days costs you not a subscription but the airway.

Context

I argued with my clinical sponsor about where patient education belonged, and I lost. Building the centralized library showed us we were both answering the wrong question: nobody had defined what the product was for. The answer was neither her shelf nor my scattered tips. Education, diet and exercise had to select for one patient — the project’s one-size-fits-one principle. 15 weeks of design and research with a practicing SLP (speech-language pathologist), handed off. Nothing was built.

Product Demo

What I threw away to stop a swallowing disorder becoming a score

I built the wireframes, the flows, the assistant, the clinical content system, the accessibility system, and the prototypes both evaluations were run on. The primary research, the concept ideation and the user testing were a team effort.

Managed at home, by the person least equipped to judge it

Dysphagia is difficulty swallowing, and getting it wrong means food or liquid entering the airway. Managing it is daily and unglamorous: exercises, a restricted diet, self-monitoring, all at home between clinic visits. Across 11 interviews, a 27-response survey and five diary studies, what kept returning was not that patients could not follow a maneuver but that they did not understand why, and the gap is different for every patient. Most are elderly, some cognitively impaired.

A line chart headed Current Dysphagia Care Fails to Adapt to Diverging Clinical Trajectories, with a y-axis labeled Level of Support and an x-axis labeled Care Timeline. A solid red Clinical Capacity curve spikes early and then falls away to almost nothing. Three dashed needs curves stay above it throughout: ALS climbs steadily, cancer falls then oscillates at a middling level, and stroke starts highest and declines. The widening area between capacity and need is shaded and labeled The Guidance Gap, where patient needs outpace provider capacity, leading to regression. Beneath, the timeline runs Pre-Diagnosis, Clinic · Acute Phase, an arrow marked Discharge · Transition Phase, and Home · Chronic Phase, with six named failures along it: understaffed and low discoverability, a diverse care journey, then generic handouts, scattered community, unverified tutorials and fragmented digital care.

The feedback was warm, and warm was the wrong temperature

Then we did the obvious thing. We are the generation that reaches for gamification, so we proposed a mascot, streaks, and a buddy. People liked it, and that nearly cost us: “that’s interesting, that’s fun”, bracketed every time by what is it, and what is it for? One SLP ended it: “If a mascot can do Mendelsohn Maneuver I would be impressed.” The feedback was neither positive nor negative. It was that we were off the point.

Headed: Dysphagia patients don’t need pressure from the streaks. They need clarity and safety. On the left, under a red cross labeled Gamification, pencil storyboards and flat mockups of a smiling water-drop character and a buddy streak list. An arrow crosses to the right, where under a green tick labeled Clear, Quantifiable Progress sit two screens: a Swallow Check scoring 21, moderate risk, one point better, above four weekly bars falling from 24 to 21; and an exercise player with an older man on video over a segmented progress bar.

A character that cannot demonstrate what it is standing beside is decoration, and decoration costs comprehension here. Cutting it took the product’s character out with it. In its place: demonstration footage of a senior the age of the patients, filmed somewhere domestic rather than clinical, and the EAT-10 where the streak had been. Both for the same reason: clarity.

The argument I lost was about the wrong thing

Education came at us from two directions: the research pointed at it, and Rinki, our clinical sponsor, asked for it outright. We proposed dissolving it into the product, bite-size, at the point of use. She pushed back hard: education was not a garnish but the point of the product, and scattering it made it something a patient could only meet by accident. I lost, and we built her library.

Headed: The answer was neither her shelf nor my scattered tips. Three panels side by side, each labeled beneath. Proposed: scattered bits, a drawn diagram of a home screen where Log a Meal, Run an Exercise and Watch a Score Move each carry their own small Explanation block. Proposed: centralized library, the same drawn home screen with an Education panel above everything else marked the whole library at the top, and the three daily actions pushed below it. Designed: tailored library, a screen capture of the Library tab open on Food, with Learn, Symptoms and Exercise beside it, filtered to My List rather than View All, listing Lemon Tart, Pumpkin Curry and Mashed Potato with the preparation each needs. The first two panels are tagged Diagram; the third is a capture.

Building it is what broke the problem open. Partway through I said it out loud to my team: “I don’t know what the app is for anymore.” One side of the file was a daily loop: log a meal, run an exercise, watch a score move. The other was a shelf of articles connected to none of it. We had been arguing about where education should sit. Neither of us had said what the app was.

The library became the second tab, opening on my symptom, my diet, my exercise: this patient’s level, not the condition’s. I never went back and told her she was wrong about centralized education, because she wasn’t. No one size fits all — only one size fits one. That idea is what the whole product is built on, and it came out of an argument I lost, not out of the research.

Nobody in the room asked what it was for

Headed: Relevance is Safety. Abundance is Noise. On the left, marked with a red cross and the words High Friction, a scatter of the earlier screens: an exercise library, a macronutrient card with carb, protein, fat and water bars, a food and drink library, a symptom picker, a serving-size stepper reading 2.0 with a millilitre-to-cup toggle, and a meal picker, all over a faded band of nutrition vocabulary reading filters, searching, serving size, cup, liter, gram, carbs, ounce. Three labeled arrows — Asking, Learning and Tracking — cross to the right, marked with a green tick and the words Low Friction. There, captioned AI Assistant, a panel greeting the patient by name and opening on her own record: a Throat Clearance insight saying food felt stuck twice this week, down from four last week, a More Insights button, and suggested questions including whether mashed potato is safe for her. Captioned My List, the Library on its Exercise tab listing three saved videos — Chin Tuck, Cough Swallow, Effortful Swallow. Captioned One Tap Logging, a Log Item screen for yogurt asking How was the swallowing?, answered Safe / No Issues, then I experienced with four symptom chips, over a Confirm button.

The product has a diet log, so we benchmarked eight diet apps on a rubric and learned from them blindly. A serving-size stepper sat inside an 11-state logging flow, and a calorie ring sat on the dashboard. Clinicians liked it. Patients liked it. In no session did anybody ask what it was for. Then a senior product designer from outside the team needed one pass: “Why am I logging portions… I thought it’s a dysphagia app.” He counted it: calories were named three times on the logging screen, the patient’s own symptom once. 11 states became six, ending on the swallow, and the ring came off.

The Outcome: One Size Fits One

Two patients on opposite trajectories need opposite things from the same app: one compensating around permanent change, the other rehabilitating toward a return. Averaging them is the failure, not the price of serving both.

Three phone screens angled on a teal-to-blue gradient: a printed restaurant menu being photographed across a table, an assistant panel returning a per-dish verdict with the modification each dish needs, and the day’s exercise sessions with the next one queued.

Daily Plan and Video Guidance: Practicing Correctly With Nobody Watching

Patients leave the clinic with a paper handout and weeks until the next visit, and a static diagram cannot specify a throat maneuver. The stack of paper became a daily checklist and a video guide performed by a real person. A session cycles a coach cue over the video once per repetition, then a timed rest, then the next maneuver. The written instructions sit alongside at body size, for anyone who cannot use the video.

Four sequential screens: the day’s home surface, the day’s exercise sessions, a video of a person demonstrating a swallow filling the top of the player screen with the coach cue set beneath it, and the exercise’s written instructions above a clinician’s note.

The Safety Loop: Prediction and Verification

Two questions decide whether a meal goes well, and standard food apps answer neither: is this safe for me before eating, and after eating, was it. The assistant and food catalog answer the first against this patient’s own IDDSI level, splitting the verdict inside a dish where it has to: on a lemon tart the smooth filling is fine, the dry crust is a major risk. It is not asked to say yes or no, but to say what to change. The diet log answers the second, every entry ending on “How was the swallowing?” What the assistant reads next week is what the patient reported this week. Calories are still the biggest number on the food’s detail card, one tap before the log. I would take that further now.

Four screens of the safety loop. A camera screen headed Scan Menu or Meal, a printed restaurant menu held up across a table. An assistant panel replying that it scanned the menu and found three options matching the patient’s L4 pureed level, each with the modification it needs — no walnuts, no fried garbanzos, no fried onions — and a safety rule warning that these dishes come with bread, to skip it as a choking hazard and ask for a spoon. A Yogurt food-detail screen carrying a Mixed Texture badge, the verdict No Prep Required with the line Matches your texture level as is, a Foods L4: Pureed chip, and a nutrition block reading 154 cal with carbs, protein and fat beside a one-cup serving selector, over a Log This Item button. And a Log Item screen for the same yogurt asking How was the swallowing?, answered Safe / No Issues, then I experienced with four chips — coughing after swallowing, acid reflux, pain when swallowing, and feeling of food sticking, which is selected — over a Confirm button.

Quantifying Recovery: Clinical and Social Visibility

Recovery between visits is invisible to the clinician, and a self-reported streak is not a measure of it. The EAT-10 is implemented straight: all 10 items, the published threshold, no score of our own. The assessment feeds a longitudinal Swallow Score and a weekly report that reaches the appointment as data rather than recollection. Beside it, a light community layer, kept because isolation was the most consistent thing patients described. A buddy compatibility score was designed and did not carry forward: peer support kept, competitive measurement cut, inside the same screen.

Four screens of one week: a longitudinal swallow score, a questionnaire item on a five-point scale, a provider-ready weekly report carrying a share action, and a message list from the peer-support surface.

Design System: Aging Vision and Color-Coded Risk

An elderly, sometimes cognitively impaired audience and a signal carried by color pull against each other, so the system is specified where they collide. Body text is Atkinson Hyperlegible Next, a low-vision typeface, at 18 and 24 pixels. Four semantic colors do two jobs at once, naming what they signal and carrying the brand.

Impact

In moderated testing with five users, all five completed onboarding and all five completed diet logging. The exercise flow succeeded for two of five: two hit an error, one was unsure. The flow that failed is the practice session above, the one I had just rebuilt around a real person instead of a character. A single blended percentage would have hidden that. The engagement closed on a full research and design handoff.

Those numbers belong to the version tested that semester. The redesign I did alone after the course has not been in front of a user: the six-state logging flow, the dashboard with the calorie ring taken off.

Everyone liked the part I had got wrong

I had put those numbers there by copying, from a category of app whose users are counting what they eat, into a product whose users are trying not to aspirate it. The mascot I caught myself, out of feedback that was warm and off-target. This one I never caught at all. It had been designed, tested and admired before somebody who had never seen the file needed one sentence to end it.

Credit

Team — Riley Liu, Raj Sureka, Emily Jeong

External partner — Dr. Rinki Varindani Desai

Special thanks — NFOSD, Ed Steger, Dr. Carrie Bruce, Kiki Marlam

Thank you also to everyone who took part in our research and evaluation sessions.